Misgastos Workflow
Automation Developer & Creator
Personal workflow for logging expenses from Telegram with n8n, Supabase, AWS Bedrock, and Metabase, without building an app.
Links
Overview
Misgastos Workflow came from a very specific need: logging expenses in seconds and regaining visibility into where my money was going. Instead of building yet another finance app, I put together a workflow where Telegram is the interface, n8n orchestrates the flow, Supabase stores the state, AWS Bedrock interprets text and images, and Metabase shows the results. I didn't design it as a public product; it's a personal automation built around how I log expenses, with Peruvian rules for Yape, Plin, merchants, categories, and payment methods.
Key features
- Expense logging from text or a screenshot sent through Telegram
- Multimodal extraction with Amazon Nova Lite via AWS Bedrock
- Real taxonomy of categories, subcategories, and merchant rules from Supabase
- Drafts with follow-up questions when data is missing before an expense is confirmed
- Expenses and state persisted in Postgres, with SQL views as AI context
- Metabase dashboard to review categories, trends, and monthly spending
- Self-hosted stack on Railway behind Cloudflare Tunnel
Screens and flows
Tech stack
Input
- Telegram Bot
- Text
- Images
Orchestration
- n8n
- Code nodes
- HTTP Request
AI
- AWS Bedrock
- Amazon Nova Lite
- Multimodal extraction
Data
- Supabase
- Postgres
- SQL views
Infrastructure
- Railway
- Cloudflare Tunnel
Analytics
- Metabase